1. Introduction

Many authors, such as Hans Primas (Reference Primas1983) and Sharon Crasnow (Reference Crasnow2000), have argued that chemistry must be treated realistically in order to establish its autonomy from physics. In this essay I consider only what has been often called “entity realism” by returning to the question of the reality of electronic orbitals. Eric Scerri contends that chemists have too long asserted a reality for electronic orbitals that is not permitted by quantum mechanics (Reference Scerri1991, Reference Scerri, Hull, Forbes and Burian1994). Originally, he (Reference Scerri1991) proposed that the continued use of orbital concepts, and their treatment as real entities, especially in educational contexts, is concomitant with textbook authors' unwarranted elevation of physics as the solution to all chemical problems. More recently, Scerri (Reference Scerri2000a, Reference Scerri2000b) claims that the use of orbitals, which are not reducible to the entities described in physics in a sense I will explain below, can strengthen the autonomy of chemistry. These “chemist's orbitals” would be treated as real entities by the chemists who use them, but need not be regarded as such by physicists and quantum chemists.

Here I will argue that the question of the “reality” of orbitals is best seen as a question about explanation, not about realism. We do not need separate entities for different disciplines as long as we remember what kind of things orbitals are: namely, mathematical wave functions. Once this point is established I will turn to reduction, and show that the question of the reality of orbitals and the question of reduction are best kept separate. The issues of realism and reduction in science are often treated with a broad brush, and I will attempt to fill in some of the finer details here. I will argue that orbitals can be treated as approximate descriptions, not referring entities, without endangering the autonomy of chemistry and chemical concepts.

2. Orbitals: Real or Not? (And in What Sense?)

An orbital is a wave function that describes the state of a single electron in an atom or molecule in terms of the electron's spatial coordinates (Levine Reference Levine2000, 134). For a single-electron system such as the hydrogen atom or the H₂⁺ ion, we can consider the orbital of the electron to describe a stationary state, because the orbital is an eigenstate of the angular momentum operator, which commutes with the Hamiltonian.Footnote ¹ However, for multielectron atoms or molecules, this commutation relationship does not hold for an individual orbital (a fact arising essentially from interelectronic repulsion), and therefore we cannot consider the electrons to be in stationary states. Because of this fact, we cannot assign indistinguishable electrons to individual “orbitals,” as is frequently done.Footnote ²

Aside from this quantum mechanical result, there are other reasons to believe that orbitals do not refer to anything actually existent. Recall that an orbital is a single-electron wave function. One of the many things such a wave function can tell us is electron density, which in turn gives an indication of the likelihood of finding an electron in a certain location upon a position measurement. In a single-electron atom, such as H or Li²⁺, the knowledge of “which orbital the electron is in,” that is, that electron's wave function, allows us to calculate the distribution of the electron density around the nucleus. Knowledge of this electron density also allows us to develop the familiar contours wherein it is, for example, 95 percent likely that the electron would be found upon measurement of its position.

For atoms with more than one electron the situation is not so simple. In order to determine the wave function for an atom of neon, for example, we must take into account not only the mutual attraction of the nucleus and the electrons, but also the repulsion that the electrons exert upon one another. A first attempt at doing so can be found in what is called the Hartree-Fock or self-consistent field method. In this approximation, the wave function of all of the electrons in the atom is assumed to be a product of the wave functions for each individual electron: Thus we assume that the wave function is separable, indicating that each electron's wave function is independent of all of the others'. A basis set of such one-electron wave functions (these could be hydrogen orbitals, but need not be) is chosen. All of the wave functions in this set, save one, are used to calculate the average potential affecting the last electron and so to determine its wave function. This new wave function is then added to the set, and this set is used to calculate the wave function for another electron. This process is repeated until new wave functions have been determined for each electron in the atom or molecule. The process is then begun anew with the new set of orbitals, and repeated until there is no significant change from one iteration of the process to the next.

At this point in the calculation we still have a total wave function that is a product of orbitals, wave functions for single electrons. However, the total energy calculated from the wave function we now have is only accurate within strict limits, due to the neglect of “electron correlation,” the instantaneous (as opposed to average) effects that the electrons have on one another: Because of their mutual repulsion, two electrons are unlikely to be near one another at a given moment. As Levine (Reference Levine2000, 315) and Scerri (Reference Scerri, Hull, Forbes and Burian1994, 164) observe, the inaccuracy in our calculations of energy, around .5 percent of the experimentally observed value, is enough to obscure the existence of an interatomic bond. For a better approximation of the system's real wave function, and hence a more accurate calculation of the total energy of the atom, we must then take electron correlation into account.

These effects are accounted for by different techniques, only one of which I will discuss here.Footnote ³ In “configuration interaction,” the electronic configuration produced by the Hartree-Fock method is supplemented with configurations corresponding to the assignment of different quantum numbers to the electrons of the atom or molecule in question. This addition of terms allows us to account for the effect the probability distribution of each electron has on the others. If an infinite number of such configurations were included, the exact wave function for the system of electrons could be produced (Berry Reference Berry1966, 296;Woody Reference Woody2000, S616). Of course, the actual approximation to the real function is better or worse depending on how many different configurations are used. The important result for our purposes is that, by adding terms to the total wave function of the atom that take account of the interaction between electrons, we, in Berry's (Reference Berry1966) terms, “spoil” most of the orbital picture of the atom.Footnote ⁴ The electron correlation effects prevent

Before we pass out of this cataclysmic moment to see what use orbitals still serve, let us pause to collect the evidence we have against a realistic interpretation of orbitals, at least in atoms with more than one electron.

In a system with a single electron, remember, the orbital is simply that electron's wave function, giving us information about energy, charge distribution, and other observables. However, if we add another electron to the system, thereby creating electron-electron interactions, we can no longer express the wave function of the system as a product of orbitals, wave functions of single electrons: “the concept of the electron orbital, a wave function which describes one electron independently of any others, is in principle incorrect if there is more than one electron” (Cohen and Bustard Reference Cohen and Bustard1966, 190). The Hartree-Fock procedure described above is an attempt to overlook this problem and express the atomic wave function as a product of orbitals. This procedure is quite successful, but it still leaves us with chemically significant margins of error. To correct this error, we must, as described above, move beyond the orbital concept altogether to an atomic wave function that, because of interelectronic interactions, cannot be expressed as a product of single-electron wave functions. Therefore, we must give up orbitals not only on the basis of the dictates of quantum mechanics but also on the basis of obtaining stricter agreement with experimental results in chemical problems. For example, as above, if we do not move beyond the orbital picture, that picture's imprecision denies us information about the possible presence of bonds.

The move away from orbitals toward greater agreement with experiment opens up a new means of evaluating metaphysical claims about the existence of orbitals. The orbitals of the Hartree-Fock method are best seen as useful approximations to the actual wave function of the atom or molecule, which itself does not include orbitals (i.e., separable wave functions for each electron). Even though a wave function expressed in terms of orbitals is not sufficiently accurate for some chemical problems, we cannot dispense with orbitals altogether.

Berry points out that Hartree-Fock orbitals are useful for predicting “charge distributions, electric field gradients at nuclei…and in certain cases…spectral line frequencies and intensities…to about the same accuracy as the total energy” (Reference Berry1966, 293). That is, these quantities can be predicted to within about .5 percent using orbitals. Woody (Reference Woody2000) emphasizes the intuitive and organizational advantages that orbitals and orbital diagrams have over wave functions generated by configuration interaction techniques: Orbital concepts capture some chemical information much more efficiently. For example, orbital models readily portray information about groups of similar molecules that a treatment with more accurate wave functions derived from configuration interaction obscures. Accurate wave functions must be developed fresh for each molecule, but orbital diagrams can be developed on the basis of shared properties. Accordingly, orbital diagrams can capture, for instance, the chemical similarities of halogens like chlorine and fluorine that are buried by more exact wave functions (Reference Woody2000, S624). In this case, accuracy of prediction of, say, energy values, is not as important as the recognition of chemical commonality.

For all their success in organizing chemical information, orbitals are an approximation to the exact wave function of the atom or molecule in question (as are the wave functions generated by the configuration interaction process), and if you rely on them for your calculations of, for example, energy, your calculations will be in error some .5 percent from the experimental results, as mentioned earlier. If you rely instead on the wave function produced by configuration interaction calculation, your results will differ from experiment by about .001 percent (Scerri Reference Scerri, Hull, Forbes and Burian1994, 164). One can also say simply that the orbitals represent the approximate distribution of charge around the nucleus, and that the wave function generated by configuration interaction calculations represents a much better approximation. If it is better agreement with the experiment one is looking for, then a wave function generated by configuration interaction seems to be the clear choice for many chemical purposes. However, orbitals can provide, as we have seen, a better look at the big picture of chemical similarity. We can thus see the two types of wave functions as providing two types of explanation: A more detailed explanation in terms of the constituents of the system and their interactions, and a less detailed, but often more useful, explanation in terms of more familiar chemical categories.

This situation is analogous to whether one considers air resistance in the theoretical consideration of projectile motion, or whether one considers intermolecular forces in the consideration of a gas: In the case of quantum chemistry, one has the choice to consider electronic correlation or not. In all three of these cases there is a compromise between empirical accuracy and simplicity of description. We can readily acknowledge that air resistance, intermolecular forces, and electronic repulsion exist, and still agree that we need to make no reference to them in the context of certain problems. If there is a metaphysical question to be asked in the case of orbitals, it is the question of what the wave function Ψ represents. This question is, of course, exceedingly difficult to answer, but the question is the same for both orbitals and more accurate wave functions; both are wave functions, plain and simple. One gives more accurate results, and the other gives a better qualitative picture, but both reflect the weird properties of quantum systems like electrons.

3. Realism and Reduction: Opposite Trends?

So far, I have said little on the relationship between the question of orbitals' existence and the question of the reduction of chemistry to quantum mechanics. It is to this question that I now turn. Scerri claims that, despite the above evidence that the wave function of an atom does not refer to individual electrons separately, the “field of chemistry continues to adhere…[to the] view of electrons in individual well-defined quantum states” (Reference Scerri1991, 317): Chemists hold electrons even in multielectron atoms to be assigned to certain orbitals. Furthermore, this view, which Scerri (Reference Scerri1991) calls the “atomic orbital” or “electronic configuration” model, has proven extremely fruitful for quantum chemistry, as mentioned above. The atomic orbital model is used as a means to classify atomic spectra, and has met with much (though not complete) success in doing so.

Scerri (Reference Scerri1991) also notes several problems with the configuration interaction technique, only one of which we need consider here. As mentioned above, the configuration interaction technique uses a linear combination of electronic configurations to approximate the correct wave function of an atom or molecule. Empirical information is usually required to truncate this linear combination, so this method of approximation is not truly from first principles. Scerri takes this last point to indicate that quantum chemistry is not reducible to quantum mechanics. He then states:

Scerri claims here that the tendency for textbooks to veil the approximate nature of orbitals indicates a reductionistic attitude: Textbooks attempt to pass off orbitals as consistent with what they take to be chemistry's ultimate quantum mechanical foundation.

The textbooks' position points to a strange state of affairs: As above, quantum mechanics denies that orbitals have any reality; orbitals are not stationary states, and we therefore cannot assign electrons to them, as Ogilvie (Reference Ogilvie1990) and Scerri (Reference Scerri2000c), have vociferously pointed out. However, Scerri (Reference Scerri1991) suggests, as we saw above, that chemistry textbooks use orbitals without stressing their approximate nature while often having a reductionistic viewpoint. These two textbook attitudes are diametrically opposed: It seems that if chemistry is using a concept, the orbital as a stationary state, denied by quantum mechanics, then it is distancing itself from quantum mechanics, not drawing nearer.

How do we make sense of the relationship between realism and reduction in this case? Crasnow suggests that the question of whether orbitals are treated realistically is intimately connected to the question of whether chemistry is reducible to quantum mechanics. She claims that the denial of the strict truth of the orbital model would present “something of a puzzle about its [the orbital model's] success and the centrality of the atomic orbital model in both the theory and practice of chemistry” (Crasnow Reference Crasnow2000, 129). Crasnow refers here to the familiar argument for realism by means of inference to the best explanation: If “theoretical constructs” like atomic and molecular orbitals do not refer to anything in external reality, how do we explain their remarkable success? On the other hand, Crasnow claims, if we do not take the option of acknowledging that the orbital model is strictly false, we are presented with two alternatives, which “may be even worse”: We can “claim that both quantum physics and chemistry should be treated realistically but that there is a radical disunity of these areas of science” (each theory refers to a “different world”) or “move to antirealism…[in which case] we give up the truth of the theories [both physics and chemistry] and there is no conflict” (Reference Crasnow2000, 129). In either case an explanation seems necessary for why there is “convergence,” in Crasnow's terms, between chemistry and physics, if they do not refer to a common world.

There is a ready solution to this dilemma. Crasnow (Reference Crasnow2000) seems to equate treating orbitals realistically and treating all of chemistry realistically; however, the latter does not imply the former. It is possible to be a realist about chemistry in general and still assert that orbitals in particular do not exactly refer to any existing thing in the world, but instead serve as devices for problem solving and visualization.Footnote ⁵ We should not judge between “the orbitals of chemistry” and “the wave functions of physics” on the basis of which is real; both orbitals and the wave functions produced by configuration interaction calculations give approximations to empirical or “real” values, and neither is strictly true; that is, neither type of wave function gives the exact wave function of a multielectron atom or molecule. Instead, in picking one of the two alternatives we need only ask ourselves how good of an approximation we need in a particular situation, whether in the classroom or in the laboratory, or what type of explanation we are looking for.

By being a realist about chemistry while acknowledging the approximate nature of orbitals, we avoid the further problem Crasnow poses:

These presuppositions are not required if we take a finer-grained stance, and allow ourselves to consider whether particular terms of chemistry refer, all the while taking a generally realistic attitude towards chemistry. One can be as much of a realist (or an instrumentalist) as one likes and still worry about whether the terms of a theory refer (or adequately capture the phenomena) precisely or only approximately. Indeed, it seems that a presupposition of realism might serve as effective a role in deciding between precision and approximation as one of instrumentalism. If we reject orbitals for particular applications, it is not because they are not real or not derivable from physics, although both of these claims can be made, as I have shown; instead, we reject orbitals in particular cases because they do not predict experimental results with an accuracy sufficient for those particular cases. In short, we reject orbitals when we do on experimental grounds, not on metaphysical grounds.

In his more recent work, Scerri has argued that regarding orbitals realistically offers a highly effective means of chemical explanation through independent chemical concepts (Reference Scerri2000a, Reference Scerri2000b). Indeed, in the interest of the autonomy of chemistry Scerri has recently proposed the use of what he calls “chemists' orbitals”:

It is unclear whether and how this proposal essentially differs from simply treating orbitals as approximations useful in certain contexts, save for Scerri's choice of ontological language. It seems Scerri (Reference Scerri2000a, Reference Scerri2000b) and Crasnow (Reference Crasnow2000) both feel a need to regard orbitals realistically for their virtues to be appreciated. However, I argue it is best to consider orbitals as approximate descriptions rather than as approximately real or occasionally real entities. The former alternative separates the metaphysical from the descriptive question, and also highlights the similarities between chemistry and physical methods of approximation.

Furthermore, recognizing that orbitals do not exactly capture the data does not imply the reduction of chemistry to quantum mechanics. Nothing prevents someone from believing that orbitals do not refer while acknowledging both their great utility and the fact that they are not derivable ab initio from quantum mechanics. To use Sarkar's criteria for reduction, it seems that the relationship between quantum mechanics and orbitals only qualifies, under the strongest possible interpretation, for “approximate abstract hierarchical reduction” (Reference Sarkar1998, 43–44). In this weak type of reduction, the only criteria met are the first two of the three: the use of rules and entities solely from quantum mechanics in the explanation of orbitals, and a hierarchical relationship between chemistry and quantum mechanics. Even the applicability of these two criteria could readily be challenged in this instance, since, as noted above, empirical information is often required in the configuration interaction calculations, which would abnegate the claim that only quantum mechanical rules are used. The hierarchical relationship between chemistry and quantum mechanics at the molecular level also deserves further examination, which I will not take up here.

What must be emphasized is that Sarkar's third criterion, the “spatial hierarchy” or part-whole relationship, is not met in the orbital case (Sarkar Reference Sarkar1998). The only spatial entities at hand are electrons and nuclei, which are at the same level of organization. It does not make sense to say that the wave functions of configuration interaction (“belonging to physics”) are physical parts of orbital wave functions (“belonging to chemistry”), or vice versa. The fact that this part-whole relationship is not met in this case strengthens the point made earlier, that orbitals are not rejected as physical entities, in what is called “entity” or “ontological eliminativism” (Reference Sarkar1998, 62–64). That is, we do not say that some wave functions are “really real” while orbitals are “mere appearance”; both are mathematical approximations with a controversial interpretation.

Furthermore, the “epistemological eliminativism” that Sarkar discusses is also avoided in the orbital case (Reference Sarkar1998, 60–62). In this process, rules and theories from the reduced realm are replaced by those from the reducing realm. This situation does not prevail in the current case: as noted above, the diagrams generated by orbital treatments have virtues that cannot be realized by more accurate wave functions. “More fundamental” quantum mechanical methods obscure valuable chemical information, but also help to indicate the range of applicability of orbital functions. Therefore both of Sarkar's (Reference Sarkar1998) arguments against epistemological eliminativism are satisfied. There is no ontological or epistemological replacement of chemistry by physics in the orbital case.

Much more could be said about realism and the relationship between physics and chemistry. Primas (Reference Primas1983) in particular argues that chemistry must use a realistic interpretation of quantum mechanics, instead of the orthodox interpretation he claims it currently employs, in order to preserve the chemists' domain of molecules and their classical and quantum properties. I have not considered the general question of the interpretation of quantum mechanics in chemistry here, but I have shown that the question of the reality of orbitals comes down to a question about the accuracy of different wave functions and of what kind of explanation is desired in a particular situation. Different interpretations will say different things about what wave functions tell us about the world, but none of them will tell us which wave functions are real and which are not.